Running nginx on a €7 box

The instinct when you spin up a new server is to over-provision. More vCPUs, more RAM, "just in case." For a static nginx origin, that instinct is almost entirely wrong.

The nginx worker process serving this page sits comfortably in tens of megabytes of resident memory. It doesn't fork per request, it doesn't hold a connection pool to a database, and it isn't compiling anything at request time. The bottleneck was never going to be compute.

What actually needed headroom was disk — the default 5 GiB volume gets tight once you add the OS, the apt cache, nginx itself, logs, and a handful of test images. Bumping to 10 GiB cost a fraction of a cent more per month and removed the only real constraint.

The lesson generalizes past this one box: before sizing a VM, ask what the workload is actually bound by. Static file serving is I/O and disk-bound, not CPU-bound. Sizing for the wrong resource means paying for capacity you'll never touch while quietly running low on the one you needed.

server {
    listen 80;
    server_name miska.blog;
    root /var/www/miska.blog;
    index index.html;
}

That's the entire server block doing the actual work. Everything else — DNS, CDN, edge logic — sits in front of it, but this is still what answers the request at the end of the chain.